In January, the technology industry gathered at CES and tackled artificial intelligence (AI) a great deal. While topics varied, some speakers focused on how AI would lend a hand to telecoms companies, even in the enterprise realm. Now, researchers have taken a step further and pointed out use cases for 6G.
A survey paper recently accepted for publication in the IEEE Open Communications Society journal delved into generative AI (GenAI) models and how they would potentially enhance the next generation of mobile connectivity.
“The integration of GenAI into 6G is an exciting new direction for enhancing and transforming the operation of these networks,” the authors said in the article. “The various GenAI models explored in this paper have been found to be useful in network administration, resource management, and performance management.”
According to the researchers, GenAI-enabled reconfigurable intelligent surfaces (RIS), digital twins, integrated sensing and communications, and unmanned aerial vehicles (UAVs) are some technologies capable of meeting the demands of future wireless networks. “However, significant challenges exist, including computational overhead, scalability, and the necessity for high-quality training data,” the paper observed.
The authors are associated with the Multimedia University (Malaysia), University of Jeddah (Saudi Arabia), Al Hussein Technical University (Jordan), Bahria University (Pakistan), Hubei University (China), and the Mohamed Bin Zayed University of Artificial Intelligence (United Arab Emirates).
Changing Networks
The paper highlights three specific use cases: environments, smart transportation, and remote healthcare monitoring.
“In dynamic network scenarios, mobility introduces significant challenges in maintaining seamless connectivity and efficient resource allocation due to the continuous movement of users, changing network topologies, and varying channel conditions,” the article reads.
Models such as Generative Adversarial Networks (GANs), the researchers said, can be crucial in addressing these challenges.
One example is how GANs enable dynamic channel estimation. The generator side creates simulated channel data under varying mobility conditions, and the discriminator side evaluates the generator’s accuracy by comparing it to real-world channel information.
“This process can be particularly valuable in environments with fluctuating signal quality, such as urban areas with dense user movement or vehicular networks.”
AI and 6G for Smart Transportation
Another area where AI may have a major impact is smart transportation, more specifically, integrated sensing and communications.
“GenAI presents a powerful solution to address these limitations [adapting to changing dynamics of traffic flow, speed, etc] by offering a framework that can adapt, predict, and optimise system behaviour in real-time,” the paper outlined.
A key AI model in this field will be the Generative Diffusion Model (GDM), the experts said.
The GDM would work through a two-step process: forward diffusion and reverse denoising.
- In the first step, Gaussian noise is progressively added to the data, creating a distorted representation.
- The model then learns to reverse this process by denoising the data through successive steps, ultimately generating realistic and accurate representations of the traffic environment.
“This ability to reconstruct data from noisy or incomplete observations is crucial in dynamic traffic scenarios where real-time information is often incomplete or corrupted due to various factors, such as interference or signal degradation,” the researchers explained.
Healthcare Monitoring
This area has long been expected to benefit from AI advancements massively. An article on the World Economic Forum website, for example, highlights that activities spanning from diagnosis to ambulance management are already changing with today’s technology.
For 6G, generative AI should help improve remote healthcare services. “The combination of IoT and GenAI offers innovative, non-contact monitoring capabilities. […] The Variational Autoencoders model (VAEs), integrated into this IoT network, can generate robust data representations by learning the underlying patterns of physiological signals and reconstructing high-quality outputs from incomplete or noisy data,” the paper authors observed.
According to them, this integration will improve diagnostic precision and optimise data transmission, enhance anomaly detection, and ensure adaptability to dynamic network conditions in remote areas.
